Looking into Crystal Balls: A Laboratory Experiment on Reputational Cheap Talk

We experimentally study cheap talk by reporters motivated by their reputation for being well informed. Evaluators assess reputation by cross checking the report with the realized state of the world. We manipulate the key drivers of misreporting incentives: the uncertainty about the state of the world and the beliefs of evaluators about the strategy of reporters. Consistent with theory, reporters are more likely to report truthfully when there is more uncertainty and when evaluators conjecture that reporters always report truthfully. However, the experiment highlights two phenomena not predicted by standard theory. First, a large fraction of reports is truthful, even when this is not a best response. Second, evaluators have diculty learning reporters' strategies and overreact to message accuracy. We show that a learning model where accuracy is erroneously taken to represent truthfulness ts well evaluators' behavior. This judgement bias reduces reporters' incentives to misreport and improves information transmission.